Skip to content
Back to skills

Workflow

ASecurity

Orchestrate multi-step workflows by decomposing and dispatching subtasks, then waiting through SAM durable parent wake delivery. Use when coordinating multiple agents or multi-phase work.

  • 60 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 2, 2026
ai-agentsreactdebuggingapi

Works with

  • api

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned October 7, 2026

npx -y skills add raphaeltm/simple-agent-manager --skill workflow --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Workflow?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Workflow
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/raphaeltm-workflow/badge)](https://www.skillsdirectory.com/skills/raphaeltm-workflow)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: workflow
description: 'Orchestrate multi-step workflows by decomposing and dispatching subtasks, then waiting through SAM durable parent wake delivery. Use when coordinating multiple agents or multi-phase work.'
---

# Workflow Orchestrator

Read the full workflow from `.claude/commands/workflow.md` and execute it.

## Quick Summary

1. **Decompose** — break the user's request into discrete subtasks with dependencies
2. **Dispatch** — send subtasks to other agents via `dispatch_task` (with `/do` instructions) and verify each task started with the intended title, profile, and constraints
3. **Wait** — persist state and a stable workflow-step `waitKey`, register `wait_for_subtasks`, and end the turn
4. **React** — dispatch dependent tasks as predecessors complete, retry failures
5. **Complete** — summarize results when all subtasks finish

## Why This Exists

SAM owns durable child-task status and can wake a sleeping parent exactly once when a registered wait resolves. The orchestrator therefore releases its prompt turn instead of maintaining an expensive foreground poller. Bounded foreground polling is a compatibility fallback only when the connected server does not advertise `wait_for_subtasks` or reports that durable wake delivery is disabled.

## Staging Debugging Access

All agents have access to `$CF_TOKEN` for direct Cloudflare API queries against staging. When monitoring subtasks that deploy to staging, use the CF API to verify their work landed correctly — query D1 for data state, read KV for feature flags, check DNS for routing. See `.claude/rules/32-cf-api-debugging.md` for the full cheat sheet.

## State Persistence

Maintain `.workflow-state.md` (local-only, snapshot-visible; never commit) as external memory. Write it before registering a wait and re-read it whenever SAM wakes the session. This survives sleep, recovery, and context compaction. See `.claude/commands/workflow.md` for the full state file format.

Files in this skill

  • SKILL.md1.9 KB
  • agents/openai.yaml293 B

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.